On Sat, 16 Feb 2013, Allin Cottrell wrote: > The doc could do with an update on this sort of thing. In > general "genr x" and "series x" are not equivalent: the former > provokes the error "No formula supplied in genr" while the > latter creates an NA series. As you know, "genr" is (sort of) > deprecated for general use; it's considered better style to > use a specific type-name instead. But... there are a few > specials that are still accomplished via "genr", namely > > genr time # linear trend > genr dummy # seasonals > genr unitdum # panel unit dummies > genr timedum # panel period dummies
On Sat, 16 Feb 2013, Allin Cottrell wrote: > I don't have any objection to that, other than the backward > incompatibility. Any others have thoughts on the matter? We could actually push the idea a little bit further: the following already works in CVS: <hansl> nulldata 24 setobs 4 1:1 q = $obsminor list D = dummify(q) setobs 6 1:1 --stacked-time-series u = $unit t = time list U = dummify(u) list T = dummify(t) </hansl> we could deprecate the "genr" special cases Allin cited above in favour of * the substitution of the "time" special thing with a dollar accessor called (well, doh) $time; * allowing dummify to take $obsminor, $unit and $time as arguments. Now this would be a rather drastic backward-incompatibility, but would make things more elegant (in the C source too). ------------------------------------------------------- Riccardo (Jack) Lucchetti Dipartimento di Scienze Economiche e Sociali (DiSES) Università Politecnica delle Marche (formerly known as Università di Ancona) r.lucchetti(a)univpm.it http://www2.econ.univpm.it/servizi/hpp/lucchetti -------------------------------------------------------